基于OCR技术的名片识别系统的研究
摘要
在实际的商务和经济活动中,名片已经成为了一个重要的身份信息载体。名片根据语言类型的不同大致可以分为两类:双语混排名片和单语名片。中英混排识别问题是亟待解决的印刷体文字识别问题,且目前的名片版面分析算法运算量大,不适用。本文试图通过研究一些新技术来推进上述问题的解决。
本文的主要工作如下:
(1)阐述名片识别系统应用的必要性,给出一般名片识别系统的总的框架图,分析名片的总体特征。
(2)论文提出基于数学形态学的版面分析方法,利用数学形态学的膨胀运算和搜索算法,实现对复杂名片版面进行快速准确的分析。
(3)针对现有字符切分方法在中英文混排环境以及不同字号文字混排情况下不能准确的进行切分等问题,提出一种基于汉字周期及识别反馈的混排文字切分方法。该方法采用基于字符间距周期的中文字符分离算法,实现对连通区域类型的判定,最后采用了基于识别的汉字部件合并算法,完成对左右结构汉字的连通区域合并。实验表明,此方法的字符切分准确率优于传统的基于投影的行字切分算法。
(4)在基于启发式规则信息分类算法基础上,利用文本在名片图像的版面位置信息来辅助分类。
(5)对论文采用的新算法进行相关实验,验证其性能。
In the actual commerce and the economic activity, the business card has already become an important status information carrier. The business card may divide into two kinds roughly according to the language type: bilingual languages and one language. The question of the mixture of Chinese and English languages is one of the questions of printing recognition. And the present business card document layout analysis algorithm’s complexity is high, no suitable. This article attempts to solve these problems through to study some new technologies.
In the paper the author makes the following contribution:
(1) The paper elaborates application necessity of the business card recognition system, gives the common the total frame chart and analyzes business card’s overall characters.
(2) The paper presents a method for card document layout analysis based on mathematical morphology. By some morphological operations and search algorithm, the proposed method can analyze a complex business card document layout quickly and accurately.
(3) Against the existing character segmentation methods can not be accurate for lines segmentation in Chinese/English mixed environment and the different size of fonts. This paper introduces a novel approach for Chinese/English mixed characters segmentation which based on periods and recognition. The method make use of the Chinese characters separation algorithm based on the character spacing cycle and achieve the determinant of the type of connective region. Finally, the algorithm completed the union of connective region of the Chinese characters by using a new Chinese character component union arithmetic based on recognition. The experiments show that this method of character segmentation accuracy is better than traditional projection based on the lines segmentation algorithm.
(4) In this paper, on the basis of heuristic rules-based information classification algorithm , we propose to use layout information in images to improve automated categorization for text information in business cards.
(5) Put up related tests for new algorithm of the paper, validating the system performance.
本文的主要工作如下:
(1)阐述名片识别系统应用的必要性,给出一般名片识别系统的总的框架图,分析名片的总体特征。
(2)论文提出基于数学形态学的版面分析方法,利用数学形态学的膨胀运算和搜索算法,实现对复杂名片版面进行快速准确的分析。
(3)针对现有字符切分方法在中英文混排环境以及不同字号文字混排情况下不能准确的进行切分等问题,提出一种基于汉字周期及识别反馈的混排文字切分方法。该方法采用基于字符间距周期的中文字符分离算法,实现对连通区域类型的判定,最后采用了基于识别的汉字部件合并算法,完成对左右结构汉字的连通区域合并。实验表明,此方法的字符切分准确率优于传统的基于投影的行字切分算法。
(4)在基于启发式规则信息分类算法基础上,利用文本在名片图像的版面位置信息来辅助分类。
(5)对论文采用的新算法进行相关实验,验证其性能。
In the actual commerce and the economic activity, the business card has already become an important status information carrier. The business card may divide into two kinds roughly according to the language type: bilingual languages and one language. The question of the mixture of Chinese and English languages is one of the questions of printing recognition. And the present business card document layout analysis algorithm’s complexity is high, no suitable. This article attempts to solve these problems through to study some new technologies.
In the paper the author makes the following contribution:
(1) The paper elaborates application necessity of the business card recognition system, gives the common the total frame chart and analyzes business card’s overall characters.
(2) The paper presents a method for card document layout analysis based on mathematical morphology. By some morphological operations and search algorithm, the proposed method can analyze a complex business card document layout quickly and accurately.
(3) Against the existing character segmentation methods can not be accurate for lines segmentation in Chinese/English mixed environment and the different size of fonts. This paper introduces a novel approach for Chinese/English mixed characters segmentation which based on periods and recognition. The method make use of the Chinese characters separation algorithm based on the character spacing cycle and achieve the determinant of the type of connective region. Finally, the algorithm completed the union of connective region of the Chinese characters by using a new Chinese character component union arithmetic based on recognition. The experiments show that this method of character segmentation accuracy is better than traditional projection based on the lines segmentation algorithm.
(4) In this paper, on the basis of heuristic rules-based information classification algorithm , we propose to use layout information in images to improve automated categorization for text information in business cards.
(5) Put up related tests for new algorithm of the paper, validating the system performance.
引文
[1] H C Fu, C S Chen, K T Sun. Recognition of Chinese Business Cards[C]. Proc.of 5th OCR & DA conference, Hshinchu,Taiwan,1996, 169-175
[2] Yaw-Huei Chiou, His-Jian Lee. Recognition of Chinese Business Cards[C]. Proc. of 4th ICDAR, Ulm,Germany,1997, 1028-1032
[3] X Lin, X Ding, Y Wu. Automatic Input System for Chinese Business Cards[C]. Proc of 7th ICCPOL, Hong Kong,1997, 277-280
[4] A Khan, S Mahmood, W MRST. A New Technique for Information Summarization [C]. Transactions on engineering, Computing and Technology V4, 2005, 1305 -1313
[5] George Nagy. Twenty Years of Document Image Analysis in PAMI[C]. IEEE Trans. On Pattern Analysis and Machine Intelligence, 2000, 22(1): 38-82
[6] K. Y. Wong, R. G .Casey, F. M, et al. Document Analysis System [C]. IBM Journal of Research and Development, 1982, 26(6):647-656
[7] T.Akiyama, N.Hagita. Automated Entry System for printed Documents[J]. Pattern Recognition,1990, 23(11):1141-1154
[8] 夏波涌.文档图像理解研究[D].中国科技大学,1998,12-16
[9] Song Mao, Tapas Kanungo. Empircal Performance Evaluation of Page Segmentation Algorithms [A]. In Proceeding of SPIE Conference on Document Recognition and Retrieval Ⅶ[C], 2000, 303-312
[10] 张利,朱颖.基于游程平滑的英文版面分割[J].电子学报,1999,27(7):26-30
[11] 周长岭.中文 OCR 中的版面分析算法初探[C].第六届全国汉字识别学术会议论文集.1996, P137-142
[12] Liody Alan Fletcher, Ranfiacher Icastur. A Robust Algorithm for Text String Separation from Mixed Text/Graphics Images[J]. IEEE Trans. PAMJ. 1988, 10(6), 910-918
[13] 刘昊.基于背景描述的中文版面分析方法的研究[D].哈尔滨工业大学,1999.6
[14] 田学东,郭宝兰.基于组合特征的中文版面分析[J].中文信息学报,1999, 13(4):22-28
[15] J M Liu ,Yuan Y, Tang. Adaptive Image Segmentation With Distributed Behavior-Based Agent[J]. IEEE Transaction on PAMI ,1997, 19(1):92-96
[16] James, Jia Li, Gio Wiederhold. Unsupervised Multisolution for Images With Low Depth of Field[J]. IEEE Trans.PAMI.2001, 23(1): 85-90
[17] Feng Z D, Huo Q. Confidence guided progressive search and fast match techniques for high performance Chinese/English OCR[A]. In: Kasturi R, Laurendeau D, Suen C, eds. ICPR 2002: the 16th Int’l Conf. on Pattern Recognition[C]. Los Alamitos: IEEE Computer Society Press, 2002,89-92
[18] Huo Q, Feng Z D. Improving Chinese/English OCR performance by using MCE-based character-pair modeling and negative training[A]. In: Antonacopoulos A, ed. ICDAR 2003: the 7th Int’l Conf. on Document Analysis and Recognition[C]. Los Alamitos: IEEE Computer Society Press, 2003, 364-368
[19] Jin J M, Wang Q R. Research on multi-language OCR systems integration[J]. Journal of Software, 2002,13:225-230
[20] Pan W M, Jin J M, Shi G S, Wang QR. A system for automatic Chinese business card recognition[A]. In: Antonacopoulos A, ed. ICDAR 2003: the 7th Int’l Conf. on Document Analysis and Recognition[C]. Los Alamitos: IEEE Computer Society Press, 2003, 1138-1141
[21] Yang Yi ming, Liu Xin. A re-examination of text categorization methods[C]. In: Proceedings of ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR), 1999, 42-49
[22] Yang Y, Pedersen J P. A comparative study on feature selection in text categorization[C]. In:Fourteenth International Conference on Machine Learning (ICML'97),1997, 412-420
[23] 张纯,张涛.中文商务名片识别系统的实现[J].中文信息学报,2002,14(2): 21-25
[24] 王彬.通讯终端上名片识别系统的实现[D].武汉理工大学,2005
[25] 张圣希,张薇,李国强,顾国庆.利用顶点链编码探测表格的斜率[J].华东师范大学(自然科学版).2004,(3):54-58
[26] Freeman H. On the encoding of arbitrary geometric configuration[J]. IRE Trans, 1961, 10 (2): 260-268
[27] Bribiesca E. A new chain code[J]. Pattern Recognition, 1999, 32: 235-251
[28] 田学东,郭宝兰. 汉字识别系统中的版面分析算法[J]. 微机发展,1999,(1):18-29
[29] 杨波,汪同庆,叶俊勇,等.文档图像的版面分析—基于数学形态学的方法[J].小型微型计算机系统,2003,9(24):68-72
[30] 刘长松,郭繁夏,丁晓青等.印刷体汉字识别方法综述[N].中国计算机报,1997,6 (23) : 86-89
[31] YI LU. Machine Printed Characters Segmentation[J]. IEEE Transactions Pattern Recognition, 1995, 28(1):67-80
[32] 杨志华,齐东旭,杨力华.基于经验模式分解的汉字字体识别方法研究[J].软件学报, 2005,16(8):1438-1444
[33] 安艳辉,田学东,郭宝兰.基于字符类别的识别反馈混排字符切分研究[J].河北大学学报,2003,23(3):55-58
[34] 陈烨,唐光荣.工业绣花软件中的字符变形算法[J].计算机工程,2004,30 (5):164-166
[35] O.Nakatnura, M.Ujiie, N. Okamoto, et al. A character segmentation algorithm formixed-mode communication[J]. System Computers Japan 16, 1985, 1277-1284
[36] Richard G Case, Eric Lecolinet. A Survey of Methods and Strategies in Character Segmentation[J]. IEEE TransaoLions on PaLLern Analysis and Machine Intelligence, 1996, 18 (7):690-706
[37] Hill,T.J. Segmentation of Postcodes for Optical Character recognition. Research Department Report 685[J], Post Office Research Centre, Martleshatn Heath, Ipswich England, 1978, 82-90
[38] Yi Lu, B. Haist, L Harrnon, et al. An accurate and efficient system for segmentation machine-printed[C]. Washington D C,text. L S Postal Sevice 5th Advanced Technology Conference, 1992, 3(11):93-105
[39] N Okatnoto, O Nakatnura, T Minatni. Character segmentation for mixed-mode cormnunicatlon in Information Processing (R. E. A Mason,Ed) [J], Elsevier ScienceNorth–holland,1983,88-96
[40] D Gellitnan, LT, Lancaster. A review of segmentation and contextual analysis techniques for text recognition Pattern Recognition[J], Pattern Recognition, 1990,(23): 335-346
[41] Kovalevsky, V A. Imagine Pattern Recognition[J]. Springer, Berlin, 1980,211-230
[42] Casey R G, Nagy G. Recursive segmentation and classification of composite character patterns[C]. Proc.6th Int’l Conf Pattern Recgnition, Munich German3, 1982, 1023-1026
[43] H Fujjsawa,Y Nakano, K Kurino. Segmentation Methods for Character Recognition: From Segmentation to Document Structure Analysis[J]. Proc.IEEE, 1992,(7):1079-1092
[44] R F H, Farage.Word-level Recognition of Cursive Script[J]. IEEE Trans, Computers, FEB, 1979,(2):172-175
[45] 李宝安,李燕,孟庆昌.中文信息处理技术—原理与应用[M].北京:清华大学出版社, 2005:120-156
[46] 周嫔,马少平,姜哲.基于单元合并地汉字切分算法的改进[J].中文信息学报,1999,(2): 3-39
[47] 赵宇明,江兴智,施鹏飞.基于笔划提取和合并地离线手写体汉字字符切分算法[J].红外与激光工程,2002,2(1):3-27
[48] 吕岳,施鹏飞,张克华.基于组件合并的手写体汉字串分割[J],软件学报,2001,11(11): 1554-1559
[49] 马少平.汉字识别系统的误识模型[J].清华大学学报,1998,(s1):108-111
[50] 陈臻刚,丁晓青,刘长松.文档识别中误切分字符拒识问题的研究[J].计算机工程与应 用, 2002,17(6):9-72
[51] Zhu X Y, Shi Y F. Feedback-Based algorithm for handwritten character recognition[J].Chinese Journal of Computers,2002,25(5):476-482
[52] 张纯,刘迎建.商务名片识别系统的设计与实现[D].中科院自动化研究所,2000,47-52
[2] Yaw-Huei Chiou, His-Jian Lee. Recognition of Chinese Business Cards[C]. Proc. of 4th ICDAR, Ulm,Germany,1997, 1028-1032
[3] X Lin, X Ding, Y Wu. Automatic Input System for Chinese Business Cards[C]. Proc of 7th ICCPOL, Hong Kong,1997, 277-280
[4] A Khan, S Mahmood, W MRST. A New Technique for Information Summarization [C]. Transactions on engineering, Computing and Technology V4, 2005, 1305 -1313
[5] George Nagy. Twenty Years of Document Image Analysis in PAMI[C]. IEEE Trans. On Pattern Analysis and Machine Intelligence, 2000, 22(1): 38-82
[6] K. Y. Wong, R. G .Casey, F. M, et al. Document Analysis System [C]. IBM Journal of Research and Development, 1982, 26(6):647-656
[7] T.Akiyama, N.Hagita. Automated Entry System for printed Documents[J]. Pattern Recognition,1990, 23(11):1141-1154
[8] 夏波涌.文档图像理解研究[D].中国科技大学,1998,12-16
[9] Song Mao, Tapas Kanungo. Empircal Performance Evaluation of Page Segmentation Algorithms [A]. In Proceeding of SPIE Conference on Document Recognition and Retrieval Ⅶ[C], 2000, 303-312
[10] 张利,朱颖.基于游程平滑的英文版面分割[J].电子学报,1999,27(7):26-30
[11] 周长岭.中文 OCR 中的版面分析算法初探[C].第六届全国汉字识别学术会议论文集.1996, P137-142
[12] Liody Alan Fletcher, Ranfiacher Icastur. A Robust Algorithm for Text String Separation from Mixed Text/Graphics Images[J]. IEEE Trans. PAMJ. 1988, 10(6), 910-918
[13] 刘昊.基于背景描述的中文版面分析方法的研究[D].哈尔滨工业大学,1999.6
[14] 田学东,郭宝兰.基于组合特征的中文版面分析[J].中文信息学报,1999, 13(4):22-28
[15] J M Liu ,Yuan Y, Tang. Adaptive Image Segmentation With Distributed Behavior-Based Agent[J]. IEEE Transaction on PAMI ,1997, 19(1):92-96
[16] James, Jia Li, Gio Wiederhold. Unsupervised Multisolution for Images With Low Depth of Field[J]. IEEE Trans.PAMI.2001, 23(1): 85-90
[17] Feng Z D, Huo Q. Confidence guided progressive search and fast match techniques for high performance Chinese/English OCR[A]. In: Kasturi R, Laurendeau D, Suen C, eds. ICPR 2002: the 16th Int’l Conf. on Pattern Recognition[C]. Los Alamitos: IEEE Computer Society Press, 2002,89-92
[18] Huo Q, Feng Z D. Improving Chinese/English OCR performance by using MCE-based character-pair modeling and negative training[A]. In: Antonacopoulos A, ed. ICDAR 2003: the 7th Int’l Conf. on Document Analysis and Recognition[C]. Los Alamitos: IEEE Computer Society Press, 2003, 364-368
[19] Jin J M, Wang Q R. Research on multi-language OCR systems integration[J]. Journal of Software, 2002,13:225-230
[20] Pan W M, Jin J M, Shi G S, Wang QR. A system for automatic Chinese business card recognition[A]. In: Antonacopoulos A, ed. ICDAR 2003: the 7th Int’l Conf. on Document Analysis and Recognition[C]. Los Alamitos: IEEE Computer Society Press, 2003, 1138-1141
[21] Yang Yi ming, Liu Xin. A re-examination of text categorization methods[C]. In: Proceedings of ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR), 1999, 42-49
[22] Yang Y, Pedersen J P. A comparative study on feature selection in text categorization[C]. In:Fourteenth International Conference on Machine Learning (ICML'97),1997, 412-420
[23] 张纯,张涛.中文商务名片识别系统的实现[J].中文信息学报,2002,14(2): 21-25
[24] 王彬.通讯终端上名片识别系统的实现[D].武汉理工大学,2005
[25] 张圣希,张薇,李国强,顾国庆.利用顶点链编码探测表格的斜率[J].华东师范大学(自然科学版).2004,(3):54-58
[26] Freeman H. On the encoding of arbitrary geometric configuration[J]. IRE Trans, 1961, 10 (2): 260-268
[27] Bribiesca E. A new chain code[J]. Pattern Recognition, 1999, 32: 235-251
[28] 田学东,郭宝兰. 汉字识别系统中的版面分析算法[J]. 微机发展,1999,(1):18-29
[29] 杨波,汪同庆,叶俊勇,等.文档图像的版面分析—基于数学形态学的方法[J].小型微型计算机系统,2003,9(24):68-72
[30] 刘长松,郭繁夏,丁晓青等.印刷体汉字识别方法综述[N].中国计算机报,1997,6 (23) : 86-89
[31] YI LU. Machine Printed Characters Segmentation[J]. IEEE Transactions Pattern Recognition, 1995, 28(1):67-80
[32] 杨志华,齐东旭,杨力华.基于经验模式分解的汉字字体识别方法研究[J].软件学报, 2005,16(8):1438-1444
[33] 安艳辉,田学东,郭宝兰.基于字符类别的识别反馈混排字符切分研究[J].河北大学学报,2003,23(3):55-58
[34] 陈烨,唐光荣.工业绣花软件中的字符变形算法[J].计算机工程,2004,30 (5):164-166
[35] O.Nakatnura, M.Ujiie, N. Okamoto, et al. A character segmentation algorithm formixed-mode communication[J]. System Computers Japan 16, 1985, 1277-1284
[36] Richard G Case, Eric Lecolinet. A Survey of Methods and Strategies in Character Segmentation[J]. IEEE TransaoLions on PaLLern Analysis and Machine Intelligence, 1996, 18 (7):690-706
[37] Hill,T.J. Segmentation of Postcodes for Optical Character recognition. Research Department Report 685[J], Post Office Research Centre, Martleshatn Heath, Ipswich England, 1978, 82-90
[38] Yi Lu, B. Haist, L Harrnon, et al. An accurate and efficient system for segmentation machine-printed[C]. Washington D C,text. L S Postal Sevice 5th Advanced Technology Conference, 1992, 3(11):93-105
[39] N Okatnoto, O Nakatnura, T Minatni. Character segmentation for mixed-mode cormnunicatlon in Information Processing (R. E. A Mason,Ed) [J], Elsevier ScienceNorth–holland,1983,88-96
[40] D Gellitnan, LT, Lancaster. A review of segmentation and contextual analysis techniques for text recognition Pattern Recognition[J], Pattern Recognition, 1990,(23): 335-346
[41] Kovalevsky, V A. Imagine Pattern Recognition[J]. Springer, Berlin, 1980,211-230
[42] Casey R G, Nagy G. Recursive segmentation and classification of composite character patterns[C]. Proc.6th Int’l Conf Pattern Recgnition, Munich German3, 1982, 1023-1026
[43] H Fujjsawa,Y Nakano, K Kurino. Segmentation Methods for Character Recognition: From Segmentation to Document Structure Analysis[J]. Proc.IEEE, 1992,(7):1079-1092
[44] R F H, Farage.Word-level Recognition of Cursive Script[J]. IEEE Trans, Computers, FEB, 1979,(2):172-175
[45] 李宝安,李燕,孟庆昌.中文信息处理技术—原理与应用[M].北京:清华大学出版社, 2005:120-156
[46] 周嫔,马少平,姜哲.基于单元合并地汉字切分算法的改进[J].中文信息学报,1999,(2): 3-39
[47] 赵宇明,江兴智,施鹏飞.基于笔划提取和合并地离线手写体汉字字符切分算法[J].红外与激光工程,2002,2(1):3-27
[48] 吕岳,施鹏飞,张克华.基于组件合并的手写体汉字串分割[J],软件学报,2001,11(11): 1554-1559
[49] 马少平.汉字识别系统的误识模型[J].清华大学学报,1998,(s1):108-111
[50] 陈臻刚,丁晓青,刘长松.文档识别中误切分字符拒识问题的研究[J].计算机工程与应 用, 2002,17(6):9-72
[51] Zhu X Y, Shi Y F. Feedback-Based algorithm for handwritten character recognition[J].Chinese Journal of Computers,2002,25(5):476-482
[52] 张纯,刘迎建.商务名片识别系统的设计与实现[D].中科院自动化研究所,2000,47-52